JP2005531853A - System and method for SNP genotype clustering - Google Patents

Abstract

A system and method for assessing genetic information and biological data applying a clustering approach, which can be used for allelic calling and genotyping. Statistical analysis of the sample data is performed at various levels to create models that relate individual data points to selected genotyping clusters and provide a relative representation of call reliability. These methods provide an integrated framework for allelic calling in various situations, and they can be applied to data obtained from various identification methods.

Description

(background)
(Field)
The present teachings generally relate to the field of genetic analysis, and more specifically to systems and methods for the analysis of biological information using a data clustering approach.

(Description of related technology)
Cluster analysis is an analysis paradigm that is frequently used to identify correlations and patterns in data. In the context of biological and genetic research, the clustering approach is based on allelic classification and gene sequence variation (insertion, deletion, restriction fragment length polymorphism ("RFLP"), short tandem repeat polymorphism (STRP) , And single nucleotide polymorphisms (including “SNPs”) for analysis purposes. In general, the clustering approach attempts to classify data points from a selected set of samples by associating data points with other data points. For example, in an exemplary SNP analysis, fluorescent probes can be used in the generation of amplification products for a large number of samples. The fluorescence values for each of the samples are quantified and then classified with respect to each other's fluorescence values by plotting the entire set of fluorescence values on a two-dimensional graph or scatter plot (scatter plot). When plotted in this manner, it can be observed that the data tend to aggregate into distinct groups according to genotype. Using this information, human observers can identify various groupings or clusters of data, and classify individual data points according to the cluster in which the data resides and select genes for the selected sample. The type can be determined.

  One of the critical limitations that hinders many conventional methods for biological data clustering analysis is that as the size of the sample set increases, the clustering analysis increasingly It takes a lot of time and effort to implement. The problem is that experimental data points cannot be easily associated with a single cluster, and as a result, the development of automated clustering tools makes them unable to analyze such data points. It worsens if it can be significantly hindered due to. To overcome these limitations, a fast, reliable, and unsupervised version for computer analysis that is capable of the level of throughput required to analyze large sample sets. It is desirable to develop a process that is). Furthermore, it is desirable to provide an analytical approach that can classify data points whose characteristics are ambiguous or difficult to characterize relative to other data points in the sample set.

(Summary)
In various embodiments, the present teachings describe systems and methods for performing allelic and genotyping by developing an underlying statistical model for cluster-based analysis. For these cluster-based analyses, incorrect information about each data point is used to determine the statistically valid cluster or class to which that data point belongs. This statistical model implements a complex analysis that can be decomposed into probabilities associated with the model itself, individual data points, and clusters formed from those data points. In general, this allelic taxonomy can operate in an unsupervised fashion that requires relatively little knowledge about the sample set apart from raw input values (eg, when no essential training data is required) .

  In one aspect, the present teachings describe a method for allelic classification that includes the following steps (a)-(d): (a) obtaining intensity information for multiple samples Wherein the intensity information includes a first intensity component associated with the first allele and a second intensity component associated with the second allele; (b) a plurality Evaluating intensity information for each of a plurality of samples to identify one or more data clusters, each of the clusters being distinct by comparing the first intensity component to a second intensity component. And (c) a likelihood model that predicts the probability that a selected sample will be present in a particular data cluster based on its intensity information Generate Step; and (d) the likelihood model applied to a plurality of samples of said step of determining the alleles constituting its associated.

  In another aspect, the teachings of the present invention describe a method for clustering analysis, which includes the following steps (a)-(d): (a) a sample comprising a plurality of data points Identifying a set, each data point having an angle value representative of the association between the first intensity component and the second intensity component; (b) a likelihood model and associated parameters Generating a set, wherein the angular value of the data point is used in determining the appropriate parameters to be used in the likelihood model, and where the likelihood model is valid Gender is evaluated by evaluating the probability that the likelihood model properly identifies the selected data point in the data set; (c) the likelihood model is determined for the data set; Okeru applied to a plurality of data points, and a step of grouping the data points to a separate cluster; step for associating the classification and selected each and its component data point and (d) a separate cluster.

  In yet another aspect, the teachings of the present invention describe a method for allelic classification, which includes the following steps (a)-(d): (a) at least two component strengths Identifying a sample set comprising each of a plurality of data points having a value; (b) evaluating a component intensity value for the plurality of data points and representing the data points as one or more representing a distinct allelic classification Grouping into data clusters; (c) generating a maximum likelihood function describing the grouping of selected data points using the component intensity values; and (d) using the maximum likelihood function. Associating each of the data points with an allelic classification.

  In another embodiment, the present teachings describe a computer readable medium. The computer-readable medium comprises the following steps: (a) obtaining experimental information about a plurality of samples, wherein the experimental information includes a first data component associated with a first allele and Including a second data component associated with the second allele; (b) evaluating experimental information for each of the plurality of samples to identify one or more data clusters comprising: Each associated with and determined in part by a combination of distinct alleles by comparing a first data component against a second data component; (c) a selected sample Generating a likelihood model that predicts the probability of existing in a particular data cluster based on the experimental information; and (d) applying the likelihood model to each of the plurality of samples. And apply to, the step of determining the alleles constituting its associated, and a command to be performed on a general purpose computer thereon.

  In yet another embodiment, the teachings of the present invention describe a computer-readable medium that includes the following steps: (a) identifying a sample set that includes a plurality of data points. Each data point has an angle value representing a relationship between the first intensity component and the second intensity component; (b) generating a likelihood model and an associated parameter set; Where the angle value of the data point is used in determining the appropriate parameters to be used in the likelihood model, and the validity of the likelihood model is Evaluated by evaluating the probability of properly identifying selected data points in; (c) the likelihood model in the sample set; Applying to the plurality of data points to group the data points into separate clusters; and (d) associating the selected classification with each of the separate clusters and its component data points into a general purpose computer. The instructions to be executed are stored on it.

  In another aspect, the teachings of the present invention describe a computer readable medium on which the following steps are performed: (a) identifying a sample set that includes a plurality of data points, the data comprising: Each of the points includes at least two component experimental values; (b) evaluating the component experimental values for the plurality of data points and grouping the data points into one or more data clusters representing distinct allelic classifications; (C) generating a maximum likelihood function describing the grouping of selected data points using the component experimental values; and (d) each of the data points using the maximum likelihood function. Instructions for causing a general-purpose computer to execute the process of associating the allele classification with each other are stored.

  In yet a further aspect, the teachings of the present invention describe a computer-based system for performing allelic classification. The system includes a database; and a program, the database for storing experimental information about a plurality of samples, the experimental information reflecting the allelic composition of each sample; the program Performs the following operations (a) to (d): (a) retrieving experimental information about a plurality of samples from the database, wherein the experimental information is the first allele Including a first data component associated with and a second data component associated with the second allele; (b) evaluating experimental information for each of the plurality of samples to identify one or more data clusters Each cluster is associated with a distinct allelic composition by comparing the first experimental component against that experimental component. And (c) a model that estimates the reliability of the likelihood model itself and evaluates how well the selected sample and its respective experimental information fit the model Generating a likelihood model that includes a fit probability assessment, wherein the model is further used to predict a probability that a selected sample is associated with a particular data cluster based on the experimental information And (d) applying the likelihood model to each of the plurality of samples to determine its associated allelic composition.

  In another embodiment, the teachings of the present invention describe a computer-based system for allelic classification. The system includes a database; and a program, the database for storing experimental information about a plurality of samples, the experimental information reflecting the allelic composition of each sample; the program Performs the following operations: (a) identifying a sample set that includes a plurality of data points, each data point having an association between a first intensity component and a second intensity component; (B) generating a likelihood model and associated parameter set, wherein the angle value of the data point is a suitable parameter used in the likelihood model; Used in determining and the validity of the likelihood model, the likelihood model was selected in the sample set (C) applying the likelihood model to a plurality of data points in the sample set and then applying the data points to separate clusters; Grouping; and (d) associating each selected cluster and its component data points with the selected classification.

(Detailed description of specific embodiments)
The teachings of the present invention describe a clustering approach that can be used to evaluate genetic information and biological data. In one aspect, these methods may be applied to a computerized analysis platform or software application, where the data analysis is performed in a substantially automated processing manner. By providing a mechanism for automated processing data analysis, the teachings of the present invention take advantage of the many limitations of conventional methods that generally require a human observer to evaluate individual data points. Handle it. Furthermore, the methods described herein can improve the speed and accuracy of analysis in large sample sets, thereby improving the efficiency of analysis in high-throughput applications.

  In various embodiments, the teachings of the present invention can also be used to evaluate sample sets that are ambiguous or difficult to classify data points. This feature is particularly useful for classifying data points that are outside or on the boundary of one or more clusters. Ambiguous data points present a significant problem in traditional clustering approaches. Because these classifications are subject to increased likelihood due to “miscalling”, which results in improper identification of data points or association with wrong clusters (ie, the data points are Actually does not belong to the cluster).

  In certain embodiments, the teachings of the present invention can be adapted to operate in conjunction with a variety of different biological and genetic data analysis applications, where clustering analysis forms a sample set. Used to solve the relationship between multiple data points. One exemplary application in which clustering analysis can be used can be used in connection with SNP positioning or identification and sample genotyping.

  SNPs represent one of several types of naturally occurring nucleotide sequence variations, and detailed SNP analysis is useful in studying the relationship between nucleotide sequence variations and diseases or other conditions It is generally thought that it can be. Currently, there are over 3 million putative SNPs identified in the human genome, and it is the goal of many researchers to validate these putative SNPs and to associate them with phenotypes and diseases. is there. One challenge in meeting this objective is that researchers need to generate and analyze large-scale genotype data that in many cases can be required to be carefully investigated and interpreted by researchers. That's what it means.

  Many analytical methods have been developed that can locate or identify SNPs. One exemplary method includes sample amplification using a pair of fluorescent probes, each probe containing a distinct marker or reporter dye specific for a different allele. During amplification, the sample is labeled according to its specific allelic composition, and the fluorescent properties of the resulting product are evaluated, and the sample is homozygous for the first allele (eg, A / A), it can be determined whether the second allele is a homozygous allele (eg A / B) or a combination of heterozygous alleles (B / B). Homozygous samples tend to display an increased degree of fluorescence in one or the other marker type, where the amount of fluorescence observed from the opposite marker is significantly reduced or completely present do not do. Conversely, a sample that is heterozygous for both alleles displays a considerable degree of fluorescence arising from both markers. A commercially available implementation of this method is the Taqman platform from Applied Biosystems, which utilizes the Prism 7700 and 7900HT sequence detection systems from Applied Biosystems to monitor and record the fluorescence of each amplified sample. To do.

  1A-D show exemplary sample sets that can be obtained according to the principles described above. Fluorescence data from amplification products for multiple samples is evaluated with respect to each other. In FIG. 1A, a scatter plot 100 can be used to visualize raw fluorescence intensity data acquired for multiple data points. In this display 100, the x-axis 105 relates to the fluorescence intensity for the first marker (red intensity) and the y-axis 110 represents the fluorescence intensity for the second marker (green intensity). Thus, each data point can be plotted with respect to other data points based on measured fluorescence intensity values.

  Allele classification of individual samples within a sample set can be achieved by evaluating the measured fluorescence value for the entire sample set with respect to another value. Visualization of exemplary data with scatter plot 100 shows that data points tend to cluster into different populations 115, 120, 125. These populations 115, 120, 125 may further be associated with a particular allelic composition or genotype, as indicated, and the first population 115 is a homozygous allelic composition of [A / A] Represents a sample with The second population 120 represents samples having a heterozygous allelic composition of [A / B]. The third population 125 represents samples having a homozygous allelic composition of [B / B].

  While the above example illustrates a sample set that forms three separate clusters, it is understood that the sample set need not meet this number alone. Thus, the sample set may contain more or fewer clusters depending on the nature and type of data being analyzed.

  There are one or more representative peripheral or outer data points 130 for the selected sample set. The observed fluorescence characteristics of this data point cannot be established very clearly with the characteristics of the main population 115, 120, 125 with which the data point 130 is associated. Using conventional analytical approaches, the appropriate allelic composition of these ambiguous or outer data points 130 can be difficult to determine with a relatively high degree of certainty or accuracy Or it may be impossible. In addition, when using traditional automated methods for clustering analyses, ambiguous data points tend to increase the frequency of renaming, and researchers can be warned to review or complete analysis Can be excluded. In various embodiments, the present teachings improve the ability to evaluate and classify ambiguous data points, thereby increasing the reliability of identification, improving automated sample identification, and reducing errors.

  FIG. 1B shows another exemplary sample set in which fluorescence intensity data is plotted as a logarithmic scatter plot 150. As shown from this graph 150, three distinct populations 155, 160, 165 corresponding to known homozygous and heterozygous alleles can be observed. Data point resolution ambiguity is further illustrated by this graph as an overlap boundary 170 between one of the homozygous population 155 and one of the heterozygous population 160. Here, each population 155, 160, 165 cannot be easily resolved, thus uniformly hindering image and automated allele recognition methods. As described in more detail hereinbelow, the present teachings assist in resolving data points in a sample set and provide a means for allelic classification and genotyping Tackle this powerful analytical problem by applying.

  In various embodiments, the grouping of data may include operations relating to the generation of prototype corners. This corner can be used to characterize and distinguish one cluster from another in a given sample set. As shown in the exemplary scatter plot 173 of FIG. 1C, each cluster of the allelic population may be associated with a distinct angular value 175, 180, 185 based on the particular characteristics of the selected cluster. . For example, the angle value 175 evaluates the average of the fluorescence intensity ratios for the data points contained within the cluster and the resulting value is the selected origin 190 in the scatter plot 173. Can be determined for homozygous clusters [A / A]. Similarly, angle values 180 and 185 may be determined in the same manner based on the corresponding heterozygous [A / B] population and homozygous [B / B] population. As described in more detail hereinbelow, the determination of the angle value represents a convenient average, whereby the data points of the sample set can be evaluated with respect to each other, and these values are It can be used as an input parameter in cluster analysis methods and subsequently manipulated during allelic classification operations.

  The determination of the angle value can also extend to each data point in the selected population, and the results can be evaluated to establish an appropriate cluster or population boundary. For example, as shown in the exemplary polarity plot 191 of FIG. 1D, the intensity value 192 for each data point can be plotted as a function of the angle value 194 to facilitate cluster analysis. Subsequently, a confidence boundary 196 may be determined based on the methods described herein to assist in associating individual data points with a particular allelic population.

  FIG. 2 illustrates a general method 200 for SNP analysis in accordance with the present teachings. In one aspect, the method 200, in state 205, includes a plurality of data points (each of which has associated component markers, ie, dye intensity values (eg, red and green fluorescence intensities)) of sample set information. Start with acquisition. The method 200 includes data obtained from a variety of different sources (eg, data obtained from dual label amplification reactions (eg, Taqman), as well as array-based detection approaches and observable properties (fluorescence, radioactivity). Can be operated in combination with other methods and other approaches designed to identify alleles based on differences). In various embodiments, each data point possesses at least two properties or characteristics (eg, two-color fluorescence), which can be used as a basis for discriminating between allelic compositions.

  After data acquisition 205, a normalization, scaling or preprocessing step 210 may be performed to modify the raw data values of the sample set as desired. This step involves correcting background fluorescence, scaling the data to a selected range, adjusting the data to fit a standard format, or a format that allows the data to be accepted for subsequent processing and analysis. Other such operations may be included.

  In one aspect, this step 210 includes a marker or dye correction routine in which intensity measurements taken for one sample or between samples are evaluated. A substantial difference between intensities may indicate that the sample data is not within the same scale and that changes between intensities may be sufficient to affect subsequent clustering analyses. In order to reduce the strong impact that substantial sample intensity differences can have on the analysis, marker or dye correction factors can be estimated and applied to the data before clustering analysis is performed.

  In addition, prior to clustering analysis, a noise correction routine can be applied to the intensity data to improve the quality of the resulting analysis. In one aspect, unwanted noise amplification can be avoided using a detection mechanism, in which data is first evaluated to determine whether a single cluster exists. In this example, certain marker or dye corrections are eliminated during the pre-processing step 210, thereby reducing an undesirable increase in noise that could otherwise adversely affect the resulting analysis. Can be avoided.

  In other embodiments, an origin normalization function may be applied during the pre-processing step 210. In one aspect, the origin normalization function uses intensity measurements associated with one or more control samples (eg, no template control-NTC). One purpose of the control sample is to provide a means for determining the background level of fluorescence for each marker or dye. Using this information, the origin normalization function can adjust the intensity values of the data to take into account the observed background. In one aspect, data normalization in this manner may be used to adjust the angular measurements of each sample depending on the location of origin. Further, if there are multiple control samples, the origin can be determined by taking the median of the control samples and adjusting the angle value for the data accordingly. Thus, if a control sample is not present or is part of a sample set, the origin normalization function may establish a reference origin to allow determination of angular measurements for each data point. In one aspect, normalized sources can be identified by searching for isolated data samples (eg, NTCs that have not been tasked) with relatively low fluorescence intensity.

  From the above description, it is understood that a number of operations can be performed on the sample set data prior to clustering analysis to improve the resulting outcome. Various approaches to data processing prior to clustering analysis (adjustment of fluorescence intensity, changes in sample data display (eg, mathematical manipulation including determination of logarithmic values and calculation of angle values), or as desired by the investigator It is envisioned that other data manipulations are possible. Thus, these operations used in combination with the clustering analysis approach described below should not be considered other embodiments of the present teachings.

  Once the sample set in state 210 has been appropriately adjusted, an ML data model is generated in state 215 based on some or all of the resulting data point values. This ML data model is a statistical model that estimates cluster model parameters using a maximum likelihood approach. In general, a separate ML data model is developed for each sample to accurately reflect the individual and unique characteristics of the selected sample set, but once a given ML data model is created It will be appreciated that it can be applied to one or more sample sets. As described in more detail below, this ML data model evaluates statistical probabilities in terms of several data points and combines the results to determine the allelic composition for each sample in the sample set. Improve existing clustering approaches by obtaining models that can be used to identify personality.

  Once the ML data model is developed, it is applied to the sample set data points in state 220 to provide a means for determining the appropriate allelic composition for the selected data points. As previously described, one desirable feature of this method 200 is that allele identification can be adapted to computer methods and requires little or no investigator input or interpretation; On the other hand, it can be done in a substantially automated fashion that keeps the accuracy of allelic calls relatively high. Thus, the results of the analysis can be output by the researcher at state 225 and other operations (eg, quality values to be generated and / or confidence scores) can be performed. The resulting information can be passed to be applied secondarily for further processing and use in the following analysis.

  In various embodiments, other data types / representations can be used with or instead of the intensity information. For example, the data used in the allele identification routine may include emission and registration data, where each signal may be characterized by a peak height and / or peak area. This information can be used in a manner similar to the intensity data to develop a likelihood model for data classification purposes.

  It is further understood that multiple methods (eg, intensity, peak height and / or peak area) can be used in combination with each other to develop a composite method for developing a likelihood model. These features can be further used to develop an independent likelihood model, which in turn can be used as a candidate likelihood model that produces improved results over other possible models. Evaluated to identify. The features used to develop the likelihood model may or may not be related to each other and can be processed / represented in a number of ways as desired by the researcher.

  In various embodiments, the data used for allelic classification can represent consensus-based values, where information corresponding to two or more data points can be combined (eg, duplicate or replicate aggregation) (Aggregation)). For example, in an array-based analysis method, multiple data points for similar sample compositions may be averaged to generate a consensus value that is then used for allelic classification according to the present disclosure. In one aspect, the aggregated data can include an associated error estimate and the outlier data can be ignored. Similarly, other statistical operations and data combinations can be contemplated for these and other analytical techniques for generating input data for allelic classification.

  In still further embodiments, the data used for allelic classification may include accompanying uncertainty, variance or tolerance information (eg, error bars or quality values). This information can be used with the underlying data that can be applied, and can be obtained in the development and evaluation of maximum likelihood equations. Further, a training data set having a known composition can be applied to a likelihood model generation method to develop a supervised method that generates and confirms an appropriate likelihood model.

  From the above, it is understood that the allele determination methods of the present teachings can be configured to operate using many different data types and data preparation methods. As a result, the use of intensity information described below (such as the input data type for an allelic classification method) should be construed as illustrative in nature and not as limiting.

  FIG. 3 illustrates a method 300 for data classification. Here, this data classification incorporates a maximum likelihood analysis approach and a model refinement routine to achieve improved allele identification. As previously described in connection with FIG. 2 above, the output information used by method 300 can include fluorescence data intensity and NTC index for each data point, which can be used to determine background. And the data intensity used in resampling can be identified. In addition, NTC information or other approaches can be used to normalize or scale the input data strength.

  In state 305, the input data is used in a model parameter estimation function, where a preliminary model is developed based on the input data to be applied to a new statistical analysis paradigm. This paradigm considers various features and hypotheses regarding allelic and genotyping. As detailed herein below, the data points of this sample set are subjected to maximum likelihood analysis. This analysis includes identifying the number of clusters present in the sample set; determining the mean, variance or standard deviation of each cluster; and estimating the frequency of this allele.

  In one aspect, the allelic classification methods of the present disclosure differ from many conventional methods for clustering analysis based on the manner in which data error or reliability estimation and propagation is handled. Unlike conventional methods that typically track error or confidence estimates and utilize this information downstream of the actual allele classification, this disclosure incorporates an error-weighted clustering approach, where The error or reliability estimate is used in clustering or grouping data by propagating this information through the classification process.

  Another distinguishing feature of the present disclosure is the application of an a priori identification approach. Here, various parameters are identified as part of the model, and a cluster model is determined that tests the model using known data values to determine whether the resulting values from the model produce the expected results. Advocated. In one aspect, an appropriate maximum likelihood equation that properly associates known data values with the model output is considered to be an appropriate equation for the following clustering analysis. Considered from another perspective, the “Apriori” model is selected in a particular putative cluster by testing individual data points against a putative cluster model and evaluating its error information. Error information can be used in cluster identification and data classification by evaluating whether the inclusion of data points produces a statistically valid result.

  Based on the “a priori” approach, model parameter estimation in state 305 is performed according to the following rules to generate an estimated maximum likelihood function.

  (1) First, consider that each data cluster in the sample set is independent of each other and each follows a single variance. This evaluation of the data produces a probability density function p (s), where the overall variance is a mixed variance defined by the following equation:

In this equation, P (C _i ) represents the “a priori” probability of each cluster, and pi (s) represents the probability density function for cluster Ci, where s is the data point of the selected sample. Indicates.

  (2) In allelic classification, each of the clusters tends to follow a binomial distribution (eg, Hardy-Weinberg equilibrium), where a relatively large population ensures minimal sampling error and an independent allele It is assumed to be accompanied by gene frequency. Assuming that the first allele “A” is “p” and the second allele “B” is “q”, generally this is (p + q) = 1 (eg, the sum of probabilities = 1) and 1−q = p.

As a result, the frequency of alleles associated with the distribution of the three clusters (two homozygous [A / A] and [B / B], and one conjugative [A / B]) is given by Can be defined by:
Formula 2: p ² (AA) + 2pq (AB) + q ² (BB) = 1
This equation shows that the probability distribution for two alleles is equal to the square of the allele probability, ie
(p (A) + q (B)) ² = p ² (AA) +2 pq (AB) + q ² (BB)
Based on the observation.

Alternatively, the probability of generating a particular allele equal to the allele frequency may be illustrated by Table 1 by way of example by Panett Square, which is expressed as p ² (AA) + 2pq (AB) + q ² It can be summed as (BB).

  (Table 1)

  (3) In calculating the angle for the data points in each cluster, the conditional Gaussian distribution follows the following formula:

  In this formula:

Indicates the average angle of the cluster C1, and σ _{i, r} indicates a parameter that is inversely proportional to the observed intensity r.

  (4) It is observed that in various sample sets, there can be outlier data points that do not tend to clearly fall into one of the identified clusters or data groupings. In one aspect, allelic classification and genotyping according to the present teachings provides a knowledge-based means for outlier detection.

  Based on the principles described above, for a selected sample set, model parameters are estimated with a likelihood function defined as a joint probability density function of data points in the sample set using a maximum likelihood (ML) criterion. This possibility function can be expressed as:

In this equation, x _n , n = 1,... N denotes all N samples, and if all samples are considered independent, the following probability function occurs:

  A maximum likelihood estimate of the parameter in state 305 can thus be obtained by maximizing the likelihood function shown above.

  Referring again to FIG. 3, identifying the appropriate parameter set in state 305, the method 300 proceeds to state 310, where data classification is based on the statistical model provided by the likelihood function. Arise. In one aspect, a Bayesian classifier approach is employed to perform an allelic call operation (eg, associating a selected data point with one of a homozygous or heterozygous cluster). Briefly described, this classifier approach uses recursive probability analysis, which probabilities the data model and determines the probability that each selected data point belongs to a cluster based on the probability model. In general, this approach applies inverse conditional logic to make an estimate as to which cluster a selected data point belongs (maximum inductive probability), and its use is described in more detail herein below. Can be modeled by a decision equation based on the following rules:

  Following the data classification in state 310, the method 300 proceeds to state 315 where confidence values are evaluated for the data points in the sample set. In various embodiments, the statistical framework for which confidence values are determined is based on a combination of several estimated statistical probabilities (eg, probability functions based on individual data point probabilities). This mode of confidence value determination distinguishes conventional methods that rely on training data sets, data models, and neural network approaches, and achieves a relatively high quality estimate of allele call confidence for each data point. During this state 315, further calculations may also be performed, including probing possible outliers and calculating an overall sample score for a selected sample set (eg, plate or array score). Including that.

  In general, confidence value determination according to the present teachings follows a joint probability analysis, where statistical evaluation is performed as a function of various experimental and analytical parameters, which are then combined to provide a confidence value for each data point. Is generated. For example, in allelic classification, confidence value determination can include combinatorial statistical analysis at the following levels: (a) a probability function or model itself, (b) a data cluster, and (c) sample data. Further details of confidence value determination are described below in combination with FIG.

  In various embodiments, the foregoing steps represent a first pass analysis of the data points of the sample set and assist in labeling and provide an initial basis for information that determines the structure or arrangement of data points relative to each other To do. In addition, this first pass analysis helps to detect outlier data points that can be identified for the purpose of reformulating the model in subsequent passes.

  Performing a preliminary or “first pass” data classification, the method 300 reaches a branch state 320 where the data may be output at state 325 or alternatively, a model. Further refinement of happens. In various embodiments, one or more “sophistication paths” can be created to refine the model used to classify the data. In general, only one refinement pass significantly improves model features and increases the overall accuracy of allelic classification for a sample set.

  Model refinement may proceed at state 330 where “outlier data” is detected. Outlier data reflects data points that generally do not fall within the limits of a single cluster and can therefore be difficult to classify. The determination of what constitutes outlier data is flexibly defined and can be based on, for example, a statistical analysis of intensity or angle values for each data point. Data points that exceed the threshold may be defined, for example, by an average value for the cluster, excluded from the analysis, and then the remaining data points may be used to define the resampling set in state 335.

  This resampling set is then used as input in state 305 to perform the next round of model parameter estimation, and the data is classified and confidence values are calculated as described above. One desirable feature of the present teachings is the ability to use the existing data points of the sample set to provide increased classification accuracy by model refinement without further training data.

  In various embodiments, for example, in array-based allelic analysis, model refinement may further include detecting or identifying NTCs that may be present (state 350). Information related to the NTC that was not previously utilized in data normalization or scaling as described above may be used in resampling in state 335. For example, NTC can be used to define a new origin from which angle measurements are made for each data point and cluster to improve classification quality.

  After a second (or third, fourth, etc.) pass data analysis, the output genotype and quality value may be distributed at state 325. In various embodiments, the output data can be stored in a database or other storage means presented to the user for examination, or redirected to another application or instrument for further post-processing. For example, the data output can be subjected to a filtering routine that identifies low quality data points, bad samples, or incorrect performance. These and other post-processing routines used in combination with the foregoing analysis methods should be considered and are other embodiments of the present teachings.

  As will be appreciated by those skilled in the art, the number of repeats used to refine the probability equation and to perform the allelic classification need not necessarily be exact. In certain situations, single pass data analysis may be sufficient to generate a good prediction quality probability equation. In other examples, possibility equation development may desirably occur over multiple iterations of the foregoing steps. Furthermore, it will be appreciated that the order of these steps may be modified as desired without departing from the scope of the present teachings. For example, the decision for model refinement 320 may precede the confidence value decision 315. In addition, other steps may be included in the method 300, for example, a data processing step including sample data integration or consensus determination occurs after data resampling 335. As a result, these and other modifications to the method for allelic determination are the present teachings and are considered alternative embodiments.

  In various embodiments, this data resampling step 335 can be used to reduce or increase the number of data points in the sample set. For example, in addition to discarding outlier data, data resampling may generate additional data points on the basis of input sample information that has passed the first iteration of probability equation determination. This approach may be weighted based on errors, uncertainties, or other information, and may skew, direct, or prefer the development of a particular type of likelihood equation or its quality.

  In one aspect, an error determination approach can be incorporated into the allele determination method, where each allelic call can be associated with a corresponding error or uncertainty value. This uncertainty value can be further determined by an error propagation method, where the uncertainty value during this allele call is monitored over one or more iterations of the likelihood equation determination. This error information may correspond to error information propagated through a theoretical error modeling process (eg, shot noise) and model fitting (eg, χ square) for the experimental cluster model used in the probability calculation.

FIG. 4 shows the stochastic component of statistical analysis 405 combined for data point evaluation. The model includes three probability components P _M 410, P _P 415, and P _C 420, where P _M 410 represents a model fit probability analysis, and P _P 415 is an inductive probability analysis for the selected cluster. It represents, and P _C 420 represents a cluster fit probability analysis on selected data points. This model fit probability P _M 410 can be used to estimate the reliability of this likelihood model itself, and generally measures how well sample points can fit this model; P _P 415 can be used to estimate the probability that a selected data point belongs to a specified allele or genotype cluster C, given an estimation model; and the in-class probability P _C 420 is Given clusters in a particular model, the selected clusters can be used to estimate the probability that a particular data point can be generated.

  The product of these probabilities can then be taken to obtain a composite probability with the specified genotype generated by the system for which the data point “s” was selected (eg, the accuracy of genotyping decisions). The joint probability). The equation representing this composite probability is given by:

Using this estimated model as a basis, the recursive probability P _P 415 can be calculated with a relatively high degree of accuracy, and the model fit probability P _M 410 and the in-class probability P _C 420 are used to define the model fit. Estimated proactively based on part. In addition, the perceived confidence value is generally related to the probability of the decision (which need not necessarily be the same), and as a result, this perceived reliability is an experimental decision probability. Note that it can be set as a function. Considered together, the composite function of probabilities forms the confidence value cv described by the equation.

  Details of the probabilities 410, 415, 420 of each component and their application to the composite analysis 405 are described in more detail herein below.

Inductive probability P _p
Inductive probability calculations generally attempt to establish what is the probability that a selected data point matches within a selected cluster relative to other clusters. As described above, a posteriori probability indicates the likelihood of the selected data points is dependent on the specific cluster based on the estimated statistical model is reflected by the condition C _J "x". If a statistical model is estimated, the inductive probability can be calculated by using a Bayesian approach. For further details on what inductive probabilities can be applied to Bayesian decision theory, the reader is referred to Duda, R .; And Hart, P .; ”Pattern Classification and Scene Analysis”; John Wiley; New York; In one aspect, the inductive probability is the following formula:

Can be determined according to

In these formulas, the a priori probability P (C _j ) can be derived from the allelic frequency by assuming a large allelic frequency p and a small allelic frequency q = 1−p. From now on, the apriori probability is:

As can be determined.

According to these formulas, P (C ₁ ) indicates the probability of having a large number of homozygous SNPs (eg, [A / A]), and P (C ₂ ) is a heterozygous SNP (eg, [A / B]). And P (C ₃ ) indicates the probability of having a small number of homozygous SNPs (eg, [B / B]).

Probability of Model Fit In one aspect, the analysis of data points can be considered from the perspective of model fit, and the application of this consideration generally affects all data points. This probability attempts to estimate what good fit is between the data point and the model. The probability of model fit can be determined by using the likelihood function as a measure of model fit, and the formula:

Defined by
In this formula, x _n , n = 1,. . . , N are surrogates for data points in the sample set. Observing the distribution of the inductive probability itself can give information about model fit, and the probability of model fit is a likelihood function and formula:

Can be defined as a function of the inductive probability that can be calculated according to or the distribution of all data points.

In-class probability _{P C}
In general, “in-class probabilities” may reflect the probability that a given data point is caused by the assigned genotype class given by the estimation model. This probabilistic analysis considers the position or location of the selected data point within the cluster (eg, the middle pair boundary of the cluster). This probability can be estimated from both the difference in angle between the point and the average angle of the model and the difference in intensity between the data point and the average intensity of the model. In one aspect, the probability estimate is the formula:

Is calculated from a separable two-dimensional Gaussian function in a defined polarity region (eg, angle-intensity region).

In formula, r is, shows data points intensity with the r _m, which shows an average model strength, theta represents the angle of the sample points with theta _m showing the average model angle, sigma _r And σ _θ denote the standard deviation for intensity and angle, respectively, and k is the scaling factor used for the confidence value scale.

  According to this formula, the first Gaussian function can be used to show the distribution of angles in the cluster with the second Gaussian function used to show the intensity distribution. Furthermore, the mean and standard deviation for intensity and angle can be calculated from the data points assigned to the clusters.

FIG. 5 illustrates an exemplary Gaussian function 500 in which the parameters for the Gaussian function are shown in angle space estimated from the data points assigned to the cluster. As previously noted, the measured standard deviation of the angle can be scaled by a selected factor to calibrate the resulting probability estimate 505. For example, the scale factor k may be set such that an angle difference of 4σ _θ yields a probability (P-value) of approximately 96.5%. Scaling in this manner can be used to include data points within 4σ _θ from the average value in the associated cluster when the confidence threshold is set to approximately 95%. It is understood that such scaling can be performed on a variety of different values to obtain different degrees of selectivity and sensitivity during data analysis. Similar Gaussian functions and scaling methods may also be applied to the intensity values for the data points of the sample set (not shown).

  From the foregoing, the methods described herein are based on allele calling and gene using a statistical model based on a clustering approach combined with knowledge from specific applications. A method for type identification is provided. These methods provide a unified framework for allelic calling in many different conditions, and various identification methodologies (eg, Taqman-based approaches, array-based identification schemes and capillary electrophoresis data (eg, , SMPlex data). In addition, confidence values derived from the various error propagation methods used to generate error estimates and the various identification methodologies described above are useful for clustering methods prior to analysis and allele calling. Can be used for input. Moreover, while the principles and structure of these methods generally remain the same for different applications, the various method parameters and thresholds seem to improve the flexibility of the method for use in other conditions. And can be adjusted according to the particular characteristics of the data used in the application.

  In addition to the analysis methods described above for model development possibilities, other model fitting methods can be used instead of or in conjunction with the allelic clustering approach. For example, chi-square fitting approaches, K-means clustering, machine learning approaches, and neural networks can be used to develop appropriate likelihood equations for data evaluation and allele determination. Moreover, the clustering confidence is assessed by using a selected likelihood model and a known sample set to assess the probability that the identified cluster features (eg, center / boundary) are acceptable. Can be done. One function of this “sanity check” is whether the selected likelihood function is associated with the selected data point and the associated allelic call with the appropriate or expected cluster. It is for evaluation.

  FIG. 6 illustrates an exemplary method 600 for array-based analysis applied to the allelic classification approach taught herein. In various embodiments, the method 600 begins at state 605 with a single registration and sample identification operation. In general, the signal associated with the array has a known location that can be associated with a particular sample composition. Thus, the signals originating from different locations on the array for the array used in the SNP analysis can each be associated with a corresponding SNP component. In one aspect, the decryption file or signal / sample identification mask can be used to create an appropriate relationship for use in analyzing the array.

  Subsequently, in state 610, the signal associated with a particular location on the array can be quantified. In certain embodiments, replication can be cohesive, and error estimates can be performed with a collection of propagating errors for further analysis.

  In state 615, an error correction routine is used that may include analysis of control signal information, expected distribution fit, normalization, and other operations designed to process the array data for further processing. Can do.

  In summary, in state 620, the aforementioned information can then be used as input, as well as used in conjunction with the aforementioned allelic classification method, and then presented to the researcher or later by other applications or instruments. Can be prepared for processing.

  FIG. 7 illustrates an exemplary system 700 that can be used to perform allelic classification according to the methods described above. In one aspect, the sample processing component 705 can provide a means for performing operations related to sample processing and data collection. These manipulations include, for example, labeling, amplifying and / or reacting the sample in the presence of a suitable marker or label; exposing the sample to a suitable analytical substrate or medium; and allelic classification Detecting a signal or radiation from a sample that is provided as input data for the method. Equipment that may be relevant to these operations includes, but is not limited to, array analysis equipment, sequencing equipment, fluorescent signal detection equipment, thermal cyclers, and other such equipment used for sample processing and data collection. Not done.

  The raw data provided by the sample processing component 705 can then be stored in the data storage component 715. This component 715 is designed for storage of data and information, including, for example, hard disk drives, tape drives, optical storage media, random access memory, read only memory, programmable flash memory devices, and other computers or electronic components. Any of various types of devices may be included. Further, data and information obtained from the sample processing component 705 may be stored and stored in a database, spreadsheet, or other suitable data structure, data storage object, or application that functions in conjunction with the data storage component 715. Can be organized.

  In various embodiments, the data analysis component 710 can be presented in the system 700. This component 710 has functionality for obtaining data and information from the sample processing component 705 or the data storage component 715. The data analysis component 710 may further provide hardware or software for performing the aforementioned allele classification method. In one aspect, the data analysis component 710 is configured to receive input data and can be stored in the data storage component 715 or displayed directly to the researcher via the display terminal 720. Processed data containing classification or genotyping information may be returned.

  Each functionality of the aforementioned components 705, 710, 715, 720 may be integrated into a single hardware device or into one or more separate devices. These devices may further have network connectivity that facilitates information transfer and data movement between devices as desired by researchers. Numerous suitable hardware and software structures that implement the allele classification methods taught herein have been developed so that each of these structures can also be considered for other embodiments taught herein. It is understood that you get.

  While the above disclosed embodiments of the present invention have shown, described and pointed out fundamental novel features of the invention as applied to the above disclosed embodiments, the illustrated apparatus, system, and / or It will be understood that various omissions, alternatives, and modifications within the detailed form of the method may be made by those skilled in the art without departing from the scope of the invention. Accordingly, the scope of the invention should not be limited to the foregoing specification but should be defined by the appended claims.

  All publications and patent applications mentioned in this specification are indicative of the level of skill of those skilled in the art to which the invention pertains. All publications and patent applications are hereby incorporated by reference to the same extent as if each respective publication or patent application was specifically and individually indicated to be incorporated by reference.

FIG. 1A is a scatter plot of the raw fluorescence intensity data obtained for a plurality of data points. FIG. 1B is an exemplary sample set with fluorescence intensity data plotted as a logarithmic scatter plot. FIG. 1C is a scatter plot in which each cluster or allelic grouping is associated with a separate angle value. FIG. 1D is an exemplary polar plot for intensity values for a plurality of data points plotted as a function of angle value. FIG. 1-2 is a continuation of FIG. 1-1. FIG. 2 is a generalized method for single nucleotide polymorphism analysis. FIG. 3 is a method for data classification that incorporates a maximum likelihood analysis approach. FIG. 4 is a block diagram illustrating components of a combined probability analysis for data classification. FIG. 5 is an exemplary angular space Gaussian function used in cluster analysis. FIG. 6 is a method for array-based analysis that incorporates a maximum likelihood analysis approach. FIG. 7 is an exemplary system for performing allelic classification.

Claims (83)

A method for allelic classification, the method comprising:
Obtaining intensity information for a plurality of samples, wherein the intensity information includes a first intensity component associated with the first allele and a second intensity component associated with the second allele. Including a process;
Evaluating intensity information for each of a plurality of samples to identify one or more data clusters, each of the clusters by comparing the first intensity component to a second intensity component; Associated with and partially determined by a combination of distinct alleles;
Generating a likelihood model that predicts the probability that a selected sample is present in a particular data cluster based on the intensity information; and applying the likelihood model to each of the plurality of samples Determining its associated allelic composition;
Including the method. The method of claim 1, wherein the likelihood model estimates a confidence in the likelihood model itself, and how much the selected samples and their respective intensity information fit the model. A method comprising a model fit probability evaluation to evaluate what to do. The method of claim 1, wherein the likelihood model includes an in-class probability estimate that estimates the probability that a selected cluster identifies a selected sample and its respective intensity information. Method. The method of claim 1, wherein the likelihood model includes an inductive probability evaluation that evaluates a probability that a selected sample and its respective intensity information belong to an assigned cluster. . 2. The method of claim 1, wherein the cluster comprises at least three distinct clusters, each data cluster being associated with a different allelic classification. 6. The method of claim 5, wherein the data cluster comprises a first cluster type associated with a first homozygous allele classification. 7. The method of claim 6, wherein the data cluster comprises a second cluster type associated with a first homozygous allelic classification. 8. The method of claim 7, wherein the data cluster comprises a third cluster type associated with a second homozygous allele classification. The method of claim 1, wherein the allelic classification is used to perform mutation analysis of one or more samples. 2. The method of claim 1, wherein the allelic classification is used to perform single nucleotide polymorphism analysis of one or more samples. 2. The method of claim 1, wherein the genotype for one or more samples is identified by performing an allelic classification. The method of claim 1, wherein intensity information for a plurality of clusters is normalized. 2. The method of claim 1, wherein the plurality of samples includes at least one "no template control" sample and associated intensity information used for sample scaling purposes. The method of claim 1, wherein the likelihood model is generated in an iterative fashion to refine the likelihood model. 15. The method of claim 14, wherein two or more iterations are used to generate a refined likelihood model. 15. A method as claimed in claim 14, wherein the refinement of the likelihood model identifies outlier samples and removes these samples, after which a refined maximum likelihood sample is generated. Implemented by generating a further likelihood model to do. 15. The method of claim 14, wherein the refinement of the likelihood model includes performing a data resampling operation, wherein a subset of the plurality of samples is the refinement. A method used to generate a generated likelihood model. 2. The method of claim 1, wherein the first intensity component and the second intensity component of the intensity information include fluorescence intensities associated with separate markers or labels. 2. The method of claim 1, wherein intensity information for each sample is obtained from a dual label amplification protocol. 20. The method of claim 19, wherein the dual label amplification protocol comprises a Taqman protocol or a SNPlex protocol. The method of claim 1, wherein the intensity information for each sample is obtained from an array-based detection protocol. A method for clustering analysis, the method comprising:
Identifying a sample set comprising a plurality of data points, each data point having an angle value representative of an association between a first intensity component and a second intensity component;
Generating a likelihood model and an associated parameter set, wherein the angle values of the data points are used in determining appropriate parameters to be used in the likelihood model, and where The effectiveness of the likelihood model is evaluated by evaluating the probability that the likelihood model properly identifies the selected data point in the data set;
Applying the likelihood model to a plurality of data points in the data set and grouping the data points into separate clusters; and each of the separate clusters and its component data points and a selected classification. A method comprising the step of associating. 23. The method of claim 22, wherein the clustering analysis is used in allelic classification. 24. The method of claim 23, wherein the allelic classification identifies distinct clusters that represent homozygous allelic classification or heterozygous allelic classification, and the identified allelic classification. Associating a data point of a particular cluster with a method. 24. The method of claim 23, wherein at least three distinct corresponding to the first homozygous allelic classification, the second homozygous allelic classification, and the first heterozygous allelic classification. There is a cluster of ways. 24. The method of claim 22, wherein the clustering analysis is used to perform a mutation analysis. 23. The method of claim 22, wherein the rastering analysis is used to perform a single nucleotide polymorphism analysis. 23. The method of claim 22, wherein the likelihood model and associated parameter set predicts the reliability of the likelihood model itself, and a selected data point defines the associated parameter set. A method that is evaluated using a probability evaluation that uses to evaluate how well the model fits. 23. The method of claim 22, wherein the likelihood model and associated parameter set predict a probability that a selected cluster will properly identify a selected data point associated with the cluster. Will be evaluated using the method. 23. The method of claim 22, wherein the likelihood model and associated parameter set use a probability estimate that predicts the probability that a selected data point belongs to a cluster into which the data point is grouped. The method to be evaluated. 23. The method of claim 22, wherein the likelihood model and associated parameter set are generated in an iterative fashion, wherein one or more data points are excluded from the model and parameter analysis, and a second A refinement model and parameter set of is generated using the remaining data points. 32. The method of claim 31, wherein the excluded data points include outlier data points that exist beyond a defined cluster threshold. 32. The method of claim 31, wherein further refinement to the model and parameter set is performed by eliminating additional data points. A method for allelic classification comprising the following:
Identifying a sample set comprising each of a plurality of data points having at least two component intensity values;
Evaluating component intensity values for the plurality of data points and grouping the data points into one or more data clusters representing distinct allelic classifications;
Generating a maximum likelihood function describing the grouping of selected data points using the component intensity values; and associating each of the data points with an allelic classification using the maximum likelihood function how to. 35. The method of claim 34, further comprising performing an assessment of confidence values for each of the data points indicating the confidence that the allele classification was made. 35. The method of claim 34, wherein at least one data point is excluded from the sample set and a refined maximum likelihood function is generated based on the remaining data points of the sample set. A method further comprising an operation. 40. The method of claim 36, wherein the excluded at least one data point comprises outlier data that resides outside the selected grouping. 35. The method of claim 34, wherein there are at least three groupings of data points, and the first homozygous allelic classification, the second homozygous allelic classification, and the first Corresponding to the heterozygous allele classification of 35. The method of claim 34, wherein the validity of the maximum likelihood function is further based on the reliability of the likelihood model itself and how well the data points fit the model. Rated, the method. 35. The method of claim 34, wherein the effectiveness of the maximum likelihood function is further evaluated according to the probability that a selected data point belongs to the associated allelic classification. 35. The method of claim 34, wherein the effectiveness of the maximum likelihood function is further evaluated according to the probability that the selected cluster is associated with a particular data point. A computer readable medium comprising the following steps:
Obtaining experimental information on a plurality of samples, wherein the experimental information includes a first data component associated with a first allele and a second data component associated with a second allele. Including a process;
Evaluating experimental information for each of the plurality of samples to identify one or more data clusters, each of the clusters partially including the first data component in relation to the second data component; By comparing and partially determined by comparing and partially determining a combination of distinct alleles;
Generating a likelihood model that predicts the probability that a selected sample is present in a particular data cluster based on the experimental information; and applying the likelihood model to each of the plurality of samples Determining its associated allelic composition;
A computer-readable storage medium having stored thereon instructions for causing a general-purpose computer to execute. 43. The computer readable medium of claim 42, wherein the first data component and the second data component include sample data intensity information. 44. The computer readable medium of claim 43, wherein the sample intensity information is obtained after reacting each of the samples using a dual label amplification protocol. 45. The computer readable medium of claim 44, wherein the dual label amplification protocol comprises a Taqman protocol or a SNPlex protocol. 43. The computer readable medium of claim 42, wherein the likelihood model evaluates the reliability of the likelihood model itself, and the selected sample and its experimental information. A computer readable medium comprising a model fit probability assessment that evaluates how well the fits the model. 43. The computer readable medium of claim 42, wherein the likelihood model estimates a probability that a selected cluster identifies a selected sample and its respective experimental information. A computer readable medium including: 43. The computer readable medium of claim 42, wherein the likelihood model comprises an inductive probability evaluation that evaluates a probability that a selected sample and its respective experimental information belong to an assigned cluster. Computer readable media including. 43. The computer readable medium of claim 42, wherein the data cluster comprises at least three distinct clusters, each cluster being associated with a different allelic classification. 50. The computer readable medium of claim 49, wherein the data cluster is a first cluster type associated with a first homozygous allelic classification, a first heterozygous allelic classification, and A computer readable medium comprising an associated second cluster type and a third cluster type associated with a second homozygous allele classification. 43. The computer readable medium of claim 42, wherein the data cluster comprises one or more clusters, each of which is associated with a distinct allelic classification. 43. The computer readable medium of claim 42, wherein the step further comprises the step of normalizing the experimental information. 43. The computer readable medium of claim 42, wherein the step further operates in an iterative fashion to refine the likelihood model. 54. The computer readable medium of claim 53, wherein the refined likelihood model is generated using two or more iterations. 55. The computer readable medium of claim 54, wherein the likelihood model is refined by identifying outlier samples and removing these samples, and then further likelihood model generation. A computer-readable medium on which is made. 43. The computer readable medium of claim 42, wherein the experimental information includes angle data. 57. The computer readable medium of claim 56, wherein the angle data is generated by comparing a first data component and a second data component for each of the data. Medium. 57. The computer readable medium of claim 56, wherein the angle data reflects a ratio of a first data component and a second data component for each of the samples. A computer readable medium comprising the following steps:
Identifying a sample set comprising a plurality of data points, each data point having an angle value representing a relationship between a first intensity component and a second intensity component;
Generating a likelihood model and an associated parameter set, wherein the angle values of the data points are used in determining appropriate parameters to be used in the likelihood model, and the likelihood The effectiveness of the degree model is evaluated by evaluating the probability that the likelihood model properly identifies the selected data point in the sample set; the plurality of likelihood models in the sample set; Instructions to cause a general-purpose computer to perform the steps of applying to the data points and grouping the data points into separate clusters; and associating the selected classification with each of the separate clusters and its component data points. A computer-readable storage medium stored above. 60. The computer readable medium of claim 59, wherein the operations are used to perform allelic classification, wherein the separate clusters are homozygous or heterozygous allelic A computer-readable medium in which a data point of a particular cluster is associated with its corresponding allelic classification. 61. The computer readable medium of claim 60, wherein the at least three distinct corresponding to the first homozygous allelic classification, the second homozygous allelic classification, and the first heterozygous allelic classification. A computer-readable medium on which there are clusters. 60. The computer readable medium of claim 59, wherein the likelihood model and associated parameter set evaluates the reliability of the likelihood model itself, and a selected data point is associated with the associated parameter. A computer readable medium that is evaluated using a probability evaluation that evaluates how well the model is fit using the set. 60. The computer readable medium of claim 59, wherein the likelihood model and associated parameter set is a probability that a selected cluster properly identifies a selected data point associated with the cluster. A computer readable medium that is evaluated using a probability evaluation to evaluate. 60. The computer readable medium of claim 59, wherein the likelihood model and associated parameter set comprises a probability evaluation that evaluates a probability that a selected data point belongs to a cluster to which it is grouped. A computer-readable medium that is evaluated using. 60. The computer readable medium of claim 59, wherein the likelihood model and associated parameter set are generated in an iterative fashion, wherein one or more data points are excluded from the model and parameter analysis. And a refined second model and parameter set is generated using the remaining data points. A computer readable medium comprising the following steps:
Identifying a sample set comprising a plurality of data points, each of the data points comprising at least two component experimental values;
Evaluating component experimental values for the plurality of data points and grouping the data points into one or more data clusters representing distinct allelic classifications;
Generating a maximum likelihood function describing a grouping of selected data points using the component experimental values; and associating each of the data points with an allele classification using the maximum likelihood function; A computer readable storage medium having stored thereon instructions to be executed by a computer. 68. The computer readable medium of claim 66, further comprising performing a confidence value assessment for each of the data points indicating the confidence that the allele classification has been made. . 68. The computer readable medium of claim 66, wherein the process further comprises a refinement operation, wherein at least one data point is excluded from the sample set and refined maximum likelihood. A computer readable medium in which a function is generated based on the remaining data points of the sample set. A computer-based system for performing allelic classification,
The system
Databases; and programs;
With
The database is a database for storing experimental information for a plurality of samples, the experimental information reflecting the allelic composition of each sample;
The program has the following operations:
Retrieving experimental information for a plurality of samples from the database, wherein the experimental information includes a first data component associated with the first allele and a second associated with the second allele. Including a data component of;
Evaluating experimental information for each of a plurality of samples to identify one or more data clusters, each cluster comprising a distinct allele by comparing a first experimental component to the experimental component; A step associated with and partially determined by the composition;
Generating a likelihood model including a reliability of the likelihood model itself and a model fit probability evaluation that evaluates how well the selected sample and its respective experimental information fits the model, The model is further used to predict the probability that a selected sample is associated with a particular data cluster based on the experimental information; and applying the likelihood model to each of the plurality of samples Performing the step of determining its associated allelic composition,
system. 70. The system of claim 69, wherein the first data component and the second data component include sample intensity data information. 70. The system of claim 69, wherein the likelihood model includes an in-class probability estimate that estimates a probability that a selected cluster identifies a selected sample and its respective experimental information. system. 70. The system of claim 69, wherein the likelihood model includes an inductive probability evaluation that evaluates the probability that a plurality of selected samples and their respective experimental information belong to an assigned cluster. ,system. 70. The system of claim 69, wherein the data cluster comprises at least three distinct clusters, each cluster being associated with a different allelic classification. 74. The system of claim 73, wherein the data cluster is a first cluster type associated with a first homozygous allelic classification, a first cluster type associated with a first heterozygous allelic classification. A system comprising two cluster types and a third cluster type associated with a second homozygous allele classification. 70. The system of claim 69, wherein the program is further operative to normalize the experimental information. 70. The system of claim 69, wherein the program further operates in an iterative fashion to refine the likelihood model. 77. The system of claim 76, wherein two or more iterations are used to generate a refined likelihood model. 77. The system of claim 76, wherein the program refines the likelihood model by identifying outlier samples and removing these samples prior to further likelihood model generation, A system that generates the refined likelihood model. 70. The system of claim 69, wherein the experimental information includes angle data generated by comparing the first data component and a second data component for each of the samples. A computer-based system for performing allelic classification,
The system includes a database; and a program;
With
The database is for storing experimental information for a plurality of samples, the experimental information reflecting the allelic composition of each sample;
The program has the following operations:
Identifying a sample set comprising a plurality of data points, each data point having an angle value indicative of an association between a first intensity component and a second intensity component;
Generating a likelihood model and an associated parameter set, wherein the angle values of the data points are used in determining appropriate parameters to be used in the likelihood model, and the likelihood Evaluating the effectiveness of the degree model by evaluating the probability that the likelihood model properly identifies the selected data point in the sample set;
Applying the likelihood model to a plurality of data points in the sample set and grouping the data points into distinct clusters; and a classification selected with each distinct cluster and its component data points A system that performs the process of associating with. 81. The system of claim 80, wherein the clustering analysis identifies distinct clusters that represent homozygous or heterozygous allelic classifications and the identified allelic classification and specific clusters. A system used in allelic classification by the process of associating with data points. 82. The system of claim 81, wherein at least three distinct corresponding to the first homozygous allelic classification, the second homozygous allelic classification, and the first heterozygous allelic classification. A system in which there are clusters. 81. The system of claim 80, wherein the likelihood model and associated parameter set predicts the reliability of the likelihood model itself, and how many selected data points are associated with it. A system that is evaluated using a probability evaluation that evaluates whether it fits the model using a parameter set.